Solutions: 15.

455x Sample Exam Questions

These sample exam problems are each worth 24 points. All sub-parts are weighted equally.

1. Suppose that an asset price St follows a lognormal, continuous-time stochastic process,

dS = µS dt + σS dB,

where µ, σ are constants and B is a standard Brownian motion. Use Itô’s lemma to
find stochastic differential equations expressing dV in terms of dt and dB for the
following functions V (S, t). Are they Itô processes?
(a) V = αS + β,
(b) V = S γ ,
(c) V = er(T −t) S,
where α, β, γ, r, and T are constants.

15.455x Page 1 of 11
 Solution: These are all Itô processes.

(a)

dV = α dS = (αµS) dt + (ασS) dBt ,

(b)

(σS)2
dV = dS γ = γ(γ − 1)S γ−2 dt + γS γ−1 dS
2
σ2

γ dS
= γS (γ − 1) dt +
2 S
2
  
σ
= γV µ + (γ − 1) dt + σ dBt ,
2
σ2
 
dV
= γ µ + (γ − 1) dt + (γσ) dBt .
V 2

(c)

dV = −rer(T −t) S dt + er(T −t) dS

 
r(T −t) dS
=e S −r dt +
S
= (µ − r)V dt + σV dBt ,
dV
= (µ − r) dt + σ dBt .
V

15.455x Page 2 of 11
 2. Let a stationary discrete-time stochastic process xt be given by

xt = A + Bxt−2 + Czt ,

where zt ∼ N (0, 1) is an IID Gaussian white-noise process, and A, B, C are constants.

(a) What is the unconditional mean of the process xt ?
(b) An analyst decides to construct a forecast fτ for future values of the process by
taking its expected value, conditional on information available up through the
time t when the forecast is made. That is,
 
fτ ≡ Et xτ | xt , xt−1 , . . . , τ > t.

Let A = 0.1, B = 0.2, C = 0.3, and suppose that two recent values x1 = 0.4,
x2 = 0.5 have just been observed. What are the one-step-ahead and two-step-
ahead forecasts? That is, at time t = 2, what are the forecasts f3 and f4 ? What
is the variance of the forecasts?

15.455x Page 3 of 11
 Solution:

(a) Taking expectations and using stationarity,

µ ≡ E [xt ] = A + BE [xt−2 ] + CE [zt ] = A + Bµ,

A
µ= .
1−B

(b) At t = 2, all returns xt and noise terms zt are unknown for t > 2 but known
for t ≤ 2. From the return equation for rt , we have

x3 = A + Bx1 + Cz3 ,
E [x3 |x1 , x2 ] = A + Bx1 = 0.1 + (0.2)(0.4) = 0.18,
x4 = A + Bx2 + Cz4 ,
E [x4 |x1 , x2 ] = A + Bx2 = 0.1 + (0.2)(0.5) = 0.2,
Var(x3 ) = E (x3 − A − Bx1 )2 = E (Cz3 )2 = C 2 = 0.09,
   

Var(x4 ) = E (x4 − A − Bx2 )2 = E (Cz4 )2 = C 2 = 0.09.

   

where the expectations are taken at t = 2.

15.455x Page 4 of 11
 3. (a) Consider the quadratic form defined by

Q(x, y) = 2x2 + 12xy − 7y 2 .

Using Lagrange multipliers, find the location and value of the extrema of Q subject
to the constraint x + 3y = 5. Determine whether each solution is a maximum,
minimum, or neither.
(b) Two assets have correlation ρ, and their volatilities are 2σ and σ respectively.
What are the weights of a minimum-variance, fully-invested portfolio of the two
assets, and what is its risk? That is, minimize the portfolio variance σp2 = w> Cw,
where C is the covariance matrix and w is an asset weight vector whose compo-
nents satisfy the budget constraint w1 + w2 = 1.
(c) In the problem above, for what values of ρ and σ will the solution also satisfy an
inequality constraint 0 ≤ wi ≤ 1? (That is, the optimal portfolio is also unlevered
and long-only.)

15.455x Page 5 of 11
 Solution:
(a) To solve along the line x + 3y = 5 using Lagrange multipliers, extremize
L = Q(x, y) − γ(x + 3y − 5)
= 2x2 + 12xy − 7y 2 − γ(x + 3y − 5).
Then differentiating,
∂L
= 4x + 12y − γ = 0,
∂x
∂L
= 12x − 14y − 3γ = 0,
∂y
∂L
= x + 3y − 5 = 0.
∂γ
Eliminating γ by subtracting the second equation from 3 times the first gives
y = 0, x=5
with solution
Q(5, 0) = 50
This single critical point is a maximum of Q along the line. Since there is only
one critical point, this can be quickly checked by evaluation Q at any other point
along the line, such as Q(−1, 2) = −50 < Qmax
Alternatively, if one substitutes y = t, x = 5 − 3t along the line, then Q =
−25t2 + 50, an unconstrained function of a single variable whose single maximum
is clearly located at t = 0.
(b) In components, we have
L = 4σ 2 w12 + σ 2 w22 + 4ρσ 2 w1 w2 − γ(w1 + w2 − 1)
which is extremized for
∂L
= 8σ 2 w1 + 4ρσ 2 w2 − γ = 0,
∂w1
∂L
= 4ρσ 2 w1 + 2σ 2 w2 − γ = 0,
∂w2
∂L
= w1 + w2 − 1 = 0.
∂γ
Subtracting the first two equations eliminates γ and gives
(8 − 4ρ)w1 + (4ρ − 2)w2 = 0,
so
4 − 2ρ
w2 = w1 .
1 − 2ρ

15.455x Page 6 of 11
 Substituting into the constraint finally gives
   
w1 1 1 − 2ρ
=
w2 5 − 4ρ 4 − 2ρ

along with

σ2
σp2 = 2 2
 
4(1 − 2ρ) + (4 − 2ρ) + 4ρ(1 − 2ρ)(4 − 2ρ)
(5 − 4ρ)2
20 − 16ρ − 20ρ2 + 16ρ3 2
= σ
(5 − 4ρ)2
4 − 4ρ2 2
= σ .
5 − 4ρ

(c) The weights wi are independent of σ. Since the correlation is bounded, −1 ≤ ρ ≤

1, w2 is always non-negative. However w1 changes sign when ρ → 1/2, so the
portfolio is unlevered and long-only provided that the correlation −1 ≤ ρ < 1/2.

15.455x Page 7 of 11
 4. The returns on a set of N assets are believed to follow the mean-reverting process

Rit − µi = −λ(Ri(t−1) − µi ) + σi zit ,

where µi , σi , λ are constants, i = 1, . . . N ; |λ| < 1; and

(
1 if t = s and i = j,
E[zit ] = 0; E[zit zjs ] =
0 if t 6= s or i =
6 j;

A market-neutral long/short trading strategy attempts to profit by investing capital

in weights assigned according to
1
wit = − (Rit − Rt ),
N

where the market average return is defined by

N
1 X
Rt = Rit .
N i=1

Assume there are no transaction costs and the risk-free rate Rf = 0. Find the expected
portfolio return
" #
X
E[Rp ] = E wi(t−1) Rit
i

in terms of the parameters given. Under what conditions is this expected return
positive?

15.455x Page 8 of 11
 Solution: We consider the more general case of using lag-k returns as the weights.
The time series of daily portfolio returns, πt is given by
X
πt (k) = wi(t−k) Rit . (1)
i∈U

This depends on the lag parameter k, so the answer should demonstrate the k-dependence
of the strategy.
Notice that the weights w depend on the the market average Rt , which in turn de-
pends on all of the stocks. In the most general case, stocks could be cross-correlated
and the result requires a full cross-covariance matrix Γk , whose (ij) matrix elements
are Cov(Rit , Rj,(t−k) ). Each Ri could be multiplied by every Rj .
In the present case, each stock is correlated only with its own lagged returns, so things
are simpler. Γk is a diagonal matrix and everything can be written as a sum of inde-
pendent AR(1) autocovariances γk (i).
The zeroth-order autocovariance is just the variance itself,
γ0 = Var[Rt ] = E (Rt − µ)2
 

= λ2 E (Rt−1 − µ)2 + E (σzt2 )

   

= λ2 γ0 + σ 2 ,
so that
σ2
γ0 = .
1 − λ2
The higher order autocovariances can be obtained by recursion.
γk = E [(Rt − µ)(Rt−k − µ)]
= −λE [(Rt−1 − µ)(Rt−k − µ)]
= −λγk−1 ,
so that
(−λ)k 2
γk = (−λ)k γ0 = σ .
1 − λ2
Now we can use this result to compute the closed-form expectation of the strategy
P/L.
N N
1−N X 1 X
E [πt (k)] = γk (i) − (µi − µ̄)2
N 2 i=1 N i=1
N
! N
(−λ)k
 
1 1 X 2 1 X
=− 1− σ − (µi − µ̄)2 .
1 − λ2 N N i=1 i N i=1
The last term, which is always negative, represents dispersion of the means. It is
independent of k and vanishes only if all the mean returns are equal. The first term
alternates in sign, so that only odd lags contribute positive P/L, and decreases in
magnitude with k.

15.455x Page 9 of 11
 5. Two stocks have prices S1 and S2 that follow geometric Brownian motion with the
same stochastic process dB:

dS1 = µ1 S1 dt + σ1 S1 dB,
dS2 = µ2 S2 dt + σ2 S2 dB.

(a) A contract has value V = S1 S2 . You can show that V also follows geometric
Brownian motion. What are its drift and volatility parameters?
(b) What is the process followed by 1/V ?
(c) A call option on V with strike K has value C(t, V ) and payoff at expiration max(S1 S2 −
K, 0). What PDE does the option satisfy?

15.455x Page 10 of 11
 Solution: It is convenient to write the series by dividing through such that the right-
hand sides are Itô processes with constant coefficients:
dS1
= µ1 dt + σ1 dB,
S1
dS2
= µ2 dt + σ2 dB.
S2

(a) Applying the product rule to V = S1 S2 ,

dV = S1 dS2 + S2 dS1 + dS1 dS2 ,

so that

dV dS1 dS2 dS1 dS2

= + + ·
V S1 S2 S1 S2
= (µ1 + µ2 ) dt + (σ1 + σ2 ) dB + σ1 σ2 ( dB)2

where we can replace ( dB)2 → dt and write

dV
= (µ1 + µ2 + σ1 σ2 ) dt + (σ1 + σ2 ) dB.
V
This standard form for geometric Brownian motion lets us read off that the drift
and volatility parameters are

µV = µ1 + µ2 + σ1 σ2 ,
σV = σ1 + σ2 .

(b) For F (t, V ) = 1/V , apply Itô’s formula:

b2 2
   
1 −1
dF = d = 2 dV + dt,
V V 2 V3
(σ1 + σ2 )2 V 2 2
 
dF dV
=− + dt
F V 2 V2
= (σ1 + σ2 )2 − (µ1 + µ2 + σ1 σ2 ) dt − (σ1 + σ2 ) dB.
 

This is also in the form of a geometric Brownian motion.

(c) Since V follows a standard geometric Brownian motion, options with V as an
underlying have values C(t, V ) that satisfy the usual Black-Scholes PDE, with
the appropriate volatility parameter for V :

∂C (σ1 + σ2 )2 2 ∂ 2 C ∂C
+ V + rV − rC = 0.
∂t 2 ∂V 2 ∂V

15.455x Page 11 of 11